Controlling Extraneous Variables Chapter5
Transcript of Controlling Extraneous Variables Chapter5
-
8/12/2019 Controlling Extraneous Variables Chapter5
1/20
ControllingExtraneousVariables
77
CHAPTER
5
A primary tenet of the experimental method is to manipulate only one variable,
keeping all other variables constant or controlled. For example, we might design
an experiment to test the role of the Nogo receptor in axonal regeneration. We
will compare axon regeneration following spinal cord injury in animals treated
with a blocker of the Nogo receptor with results from animals not treated with the
receptor blocker. If the groups differ, we conclude that the Nogo receptor playsa role in axon regeneration.
WHAT VARIABLES NEED TO BE CONTROLLED?
Must every variable be controlled? You will need to make a careful judgment as
you design your experiment. The beauty of the experiment is that you can con-
clude that your independent variable caused the difference in the dependent vari-
ableas long as no other variables differed across the groups. When you design
an experiment, you will realize it can be very difficult to ensure that absolutely
no other variable changed across the groups. If animals were tested on different
days of the week, that is an uncontrolled variable. What about the identity of the
person testing the animals? You can begin to see that the experimenter must
worry about controlling as many variables as possible, but that some variables
-
8/12/2019 Controlling Extraneous Variables Chapter5
2/20
are probably more important than others, depending on your particular experi-
ment. It is helpful to study papers reporting other experiments on the same topic.
Variables important to control in the study of circadian rhythms (for example,
the lighting of the animals housing room) are different from variables important
to control in the study of development of motor learning (for example, priormotor experience).
CONTROLLING SUBJECT VARIABLES
Subject variables are variables associated with each subject or participant in your
experiment, such as the age, prior experience, or body weight of the subject. As
discussed in Chapter 4, random assignment is a powerful tool for controlling vari-
ables. If you randomly assign subjects to groups, you will distribute the multitude
of subject variables you did not explicitly control across those groups. With a large
enough sample size, you can be assured that there is no reason to suspect thegroups differ on those characteristics. Of course, in most research situations with
laboratory animals, you can select subjects of a particular gender and age, with a
particular range of prior experience, perhaps from a specially inbred population.
Be aware that even genetically identical mice can show individual differences. A
fetus positioned next to two males in utero receives different hormone exposure
than one next to two females, and this influences later behavior (for a review, see
Ryan & Vandenbergh, 2002). Many other pre- and post-natal factors can lead to
individual differences (Lathe, 2004) or can make one group of mice behave in a
different manner from another. When your experiment depends on human par-
ticipants, you will be even more concerned about controlling subject variables.
One approach is to explicitly control for as many variables as you can and thenuse random assignment to take care of the rest.
Within-Subjects Design
In some instances, the experimenter might want to control for variables
associated with the subject by conducting an experiment using the within-
subjects design. In the within-subjects design, each subject of the experiment
experiences all the experimental conditions. For example, if the design is simple,
with one independent variable and one control condition, then in a within-
subjects design, each subject experiences both the independent variable and the
control condition.Let us say you are interested in the effect of a new anesthetic on the sensitiv-
ity of GABA receptors to the neurotransmitter GABA. This is of interest because
the GABA receptor seems to play a crucial role in the loss of consciousness
78 The Design of Experiments in Neuroscience
-
8/12/2019 Controlling Extraneous Variables Chapter5
3/20
associated with the effect of anesthesia. You might approach this question by mea-
suring a response such as the whole cell current evoked by GABA application,
reflecting the passage of ions through the GABA receptor. You could construct a
dose-response curve for GABA application either with or without anesthetic. You
can measure this response from Xenopus oocytes (frog eggs) that have beeninjected with mRNA encoding for GABA receptor subunits. Researchers use this
approach in part because these oocytes are large and sturdy, providing an easily
manipulated cell for expression of the GABA receptors of interest. Their mem-
branes are also not studded with a multitude of other receptors, keeping the exper-
imental situation simple. Because each oocyte might differ slightly in terms of the
expression of GABA receptors, a within-subjects design is wise. You could mea-
sure a dose-response curve for GABA action for each oocyte in the presence and
in the absence of anesthetic.
Experimenters often employ the within-subjects design in studies measuring
levels of brain activity using a technique such as fMRI (functional magnetic res-
onance imaging). The between-subjects variability in fMRI is so great that anexperimenter gains a major advantage if a within-subjects design can be used
(Huettel, Song, & McCarthy, 2008).
Matched-Samples Design
In amatched-samples design,you control for a variable associated with the
experimental units by measuring that variable and then matching subjects in
the experimental and control group for that variable. For example, in studying the
effects of practice on the brain areas activated during a mental imagery task, you
might assign participants to groups based on their initial ability to complete the
mental imagery task. People vary widely in native ability for mental imagery tasks,and thus this is an important variable to control. You could randomly assign par-
ticipants to groups and hope that mental imagery ability was equally distributed
across the groups, but in reality this works best only for large samples; in some
cases it is impossible to randomize. If you have a small sample, and can measure
this important variable in a pretest, then it is best to use a matched-samples design.
To match the samples, first administer a pretest to measure mental imagery abil-
ity. Then order the people from first to last according to their score on the pretest.
If your experiment consists of two groups, an experimental and a control group,
you can take the first two top-scorers on the pretest and randomly assign one to
the experimental group and one to the control group. Proceeding down the list,
two people at a time, you could fill your two groups with people matched on men-tal imagery ability. Now you could conduct your study with two groups, know-
ing that the experimental and control groups were matched in terms of mental
imagery ability (see Figure 5.1).
Chapter 5 Controlling Extraneous Variables 79
-
8/12/2019 Controlling Extraneous Variables Chapter5
4/20
Attrition
Attrition is the loss of subjects before or during your experiment. You can ran-
domly assign who begins a study, but you have little control over who completes it.
Selective attrition in one experimental group can raise concerns about confounding
variables. Might the subjects who dropped out differ in some way from those who
remained? When you design your study, you should be concerned if one condition
is somehow less attractive to participants and might encourage a higher dropout ratethan the other conditions do. For example, if you design a study on the effects of
sleep deprivation on learning, participants assigned to the condition of having only
2 hours of sleep during the night might drop out at a higher rate than those assigned
to the condition of having 8 hours of sleep. You might be left with participants in
80 The Design of Experiments in Neuroscience
Figure 5.1 Schematic Showing the Matched-Samples Design
1
Rank orderingon premeasure
All animals
receive
dependent
measure
(neurochemical
analysis of
brain tissue)
Compare
results of
experimental
and control
animals on
neurochemical
analysis
Rank order of
monkeys in
blood chemistry
C
F
A
J
H
D
E
GI
B
2
Formationof pairs
Paired
monkeys
CF
AJ
HD
EG
IB
3
Randomization
4
Treatment
Experimental group
F, A, H, E, I
Control group
C, J, D, G, B
Experimental group
F, A, H, E, I
Control group
C, J, D, G, B
65
SOURCE: From Ray,Methods Toward a Science of Behavior and Experience,7E. 2003 Wadsworth, a part of Cengage
Learning, Inc. Reproduced by permission. www.cengage.com/permissions
-
8/12/2019 Controlling Extraneous Variables Chapter5
5/20
-
8/12/2019 Controlling Extraneous Variables Chapter5
6/20
intensive interventions for reading disabilities, instead of using a control group of
children receiving only standard interventions.
And what were the results? The phonologically based intervention did
increase activity in left hemisphere brain regions of interest, and these increases in
activity were evident both immediately after intervention and also at the 1-yearfollow-up. Will future teachers take brain scans to determine how well their
students are doing? Based on this study, some already are.
WHEN A SUBJECT VARIABLE IS YOUR
INDEPENDENT VARIABLE: QUASI-EXPERIMENTS
When a subject characteristic is your independent variable, you can often design a
study that is otherwise like a true experiment. You can control other variables care-
fully and measure your dependent variable in an unbiased and objective manner. Of
course, you are usually unable to randomly assign subjects to groups, so youaremiss-ing a key quality of a true experiment, making this a quasi-experiment. For example,
the investigator may be interested in the differences between the brains of people
with multiple sclerosis and the brains of healthy controls. Obviously, this researcher
cannot begin the experiment by randomly assigning participants to either the mul-
tiple sclerosis or control condition. This is an example of a research question that
only a quasi-experiment can address. You can apply methods to control for other
subject variables that might be relevant, such as the matched-samples design. For
example, in a study comparing patients with multiple sclerosis to healthy controls,
you could match participants on age, gender, other health variables, and so on.
The inability to randomly assign participants to groups poses important
problems for interpreting results. Even if you have all other important variablescarefully controlled, your results could be due to multiple confounding variables.
Your two groups may differ not just in the diagnosis of multiple sclerosis. They
also may differ in lifestyle factors, drug history, and ability to care for their own
general health. Anything correlated with the experience of multiple sclerosis in
our current society will be different between your two groups. If you find dif-
ferences between the groups in your dependent measures, such as different levels
of brain activity in specific regions, how can you determine if multiple sclerosis
causes this difference or if a variable associated with multiple sclerosis causes it?
Transgenic or Knockout Mice
An increasingly common technique is to use transgenic, knockoutor knockin,
mice to investigate an independent variable. Atransgenicmouse has a segment of
artificially constructed DNA (atransgene) incorporated into its genome that may
cause that mouse to produce the protein coded for by the transgene. A knockin
mouse has a gene inserted at a targeted site in the genome, allowing over-expression
82 The Design of Experiments in Neuroscience
-
8/12/2019 Controlling Extraneous Variables Chapter5
7/20
of the protein product of that gene. A knockoutmouse has a particular gene inac-
tivated or disrupted, so the protein product of that gene is no longer produced or is
produced in a truncated form and thus is no longer functional (see Figure 5.2).
Control mice for such an experiment are called wildtype mice, and they are best cho-
sen from littermates of the mutant mice. Genotyping is used to determine which micehave the genetic alteration and which are genetically normal.
If you are comparing mice with a mutation in, say, the sodium channel gene,
with control mice, is this a true experiment or a quasi-experiment? Generally,
experimenters deliberately produced mice with the specific gene mutation,
thereby giving the appearance of this being a manipulated variable. However,
the independent variable is a subject variable and thus this is a quasi-experiment.
Many of the studies currently published in leading neuroscience journals are
quasi-experiments, comparing such genetically altered mice to controls.
What are some of the problems in making interpretations? Differences between
mutated and control groups might be attributed to the mutated gene, or they might
Chapter 5 Controlling Extraneous Variables 83
Figure 5.2 A Conditional Knockout
Brain sections showing location of mRNA for cannabinoid 1 receptors (CB1; shown in black) for wildtype mice (WT) and
for three different types of conditional knockout mice. The CaMK-CB1-/- mice lack CB1 in all principal neurons but main-
tain CB1 in GABA-containing cells (scattered dots in E, F, and H). The glu-CB1-/- mice lack CB1 in cortical cells that con-
tain the neurotransmitter glutamate.The GABA-CB1-/- mice lack CB1 in cells using the neurotransmitter GABA.
SOURCE: Monory et al., 2006.
-
8/12/2019 Controlling Extraneous Variables Chapter5
8/20
be due to the myriad compensatory changes induced in a system developing and
functioning without that gene. A change in one gene can cause a cascade of changes
in a developing system, so by the time you study the adult you are looking at the net
effect of many variables, some directly attributable to your transgene and others
being more indirect consequences (see Box 5.1). You can apply some of the designprinciples we have discussed, such as using a within-subject or a matched-sample
design, to strengthen your experiment. One important control group specific to this
approach is that of littermates without the transgene. Littermate controls can help
determine if the transgene alters offspring by effects on the mother, effects experi-
enced either prior to birth or before weaning.
These difficulties are what spur researchers to develop ways to produce
conditional gene knockouts: ways to control the activation of a gene by various
conditions. You might use a spatial or a temporal conditional knockout. A spatial
conditional knockout would have the gene inactivated only in certain brain
regions, whereas a temporal conditional knockout would have the gene inactivated
only at certain times in development. The inactivated gene can be controlled by asubstance fed to the animal in some cases or by a regulatory region that restricts
expression only to certain cell types (for more details see Gaveriaux-Ruff &
Kieffer, 2007). With a conditional knockout, a better controlled experiment is pos-
sible. One group of your conditional knockout mice can be treated so as to disrupt
function of your gene of interest at a specific time or in a specific brain region, while
a second group of the conditional knockout mice could serve as controls. For
example, researchers were interested in identifying where in the brain marijuana
acted to produce its effects (Monory et al., 2007). Marijuana acts largely through
the CB1 receptor and, while this is most densely located on GABA-containing cells
in the cerebral cortex, it is also found on cells in many other brain areas. Using var-
ious strains of mice with the CB1 receptor gene inactivated in different cell popu-lations, they were able to find that different effects of marijuana (e.g., effects on
movement, pain response, or body temperature) were mediated by different brain
regions. Surprisingly, the cortical GABAergic neurons with high levels of CB1
receptors did not seem important in mediating these responses.
Researchers are aware of these common interpretational difficulties and gen-
erally consider studies of transgenic mice as one portion of a wider series of stud-
ies furnishing convergent evidence for a hypothesis. For example, transgenic mice
with disruption of the neuropeptide Y gene show alterations in common measures
of anxiety in mice (see Figure 5.3), and researchers consider this evidence in the
context of other studies, for example, using drugs to alter neuropeptide Y levels
to determine if we can support a strong link between anxiety and this peptide.
Each technique has some drawbacks, but when a new technique provides further
support for a hypothesized link between neuropeptide Y and anxiety, researchers
may focus less on each specific techniques drawbacks.
84 The Design of Experiments in Neuroscience
-
8/12/2019 Controlling Extraneous Variables Chapter5
9/20
-
8/12/2019 Controlling Extraneous Variables Chapter5
10/20
PLACEBO CONTROLS
The use of a placebo control is common in studies using drug administration in
humans. Placebos (from the Latin to please) were originally inert pills that
physicians gave to patients with a malady for which there was no other treatment.Surprisingly, there is a measurable improvement from a placebo treatment in many
cases. To what should we attribute this improvement? There are probably many
causes. Sometimes a persons health improves simply with the passage of time, an
event referred to as an unexplained remission. On the other hand, some studies
have included groups with no placebo and compared these to groups given a
placebo; if there is greater improvement in the placebo-treated group, the differ-
ence in improvement cannot be attributed to unexplained remission. Sometimes
the attention from a physician and the sense of being treated will buoy a person
up and, through a still poorly understood process, lead to improvement in health.
Placebos are more effective the more dramatic they are; larger pills are more effec-
tive than smaller ones, surgical interventions more effective than less invasiveapproaches. Expectation is an important factor as well; if both the patient and the
doctor express an expectation that the treatment will be effective, the placebo
effect is larger. There is an interesting literature on the neurobiological effects of
placebos (for example, see Price, Finniss, & Benedetti, 2008). For example,
patients with Parkinsons disease are often treated with medication that increases
dopamine release or implanted electrodes in the subthalamic nucleus. Studies show
that placebo treatments can have beneficial effects on motor symptoms. These
beneficial placebo effects are actually accompanied by increased dopamine release
or changes in firing patterns of subthalamic nucleus neurons.
In some cases, it is difficult to control for the placebo effect. Studies of seasonal
affective disorder (i.e., the winter blues, a depressed mood that regularly occurs
on a seasonal basis) have suggested that some people with this disorder show lifted
moods if they expose themselves to bright lights for several hours a day. Circadian
rhythms researchers found this unsurprising because the decrease in the number of
hours of daylight experienced by organisms living far from the equator triggers sea-
sonal responses. In fact, circadian researchers predicted that treatment with several
hours of light in the morning would be more effective than the same light treatment
given midday. Did the results support this prediction that morning light would be
most effective? Several studies found some support for this, but the effect was not
overwhelming. What might be happening? One possibility is that people were expe-
riencing a placebo effect. Disrupting your daily routine to sit in front of a light box
for several hours may lift your mood by the mere sense of being under treatment.
How could researchers control for this effect? Asking people to sit in front of dim
lights is one possibility, but this might be less effective as a placebo as well as less
effective as a light stimulus. One study used sham negative ion generators as a
placebo control group for bright light therapy (Eastman, Young, Fogg, Liu, &
86 The Design of Experiments in Neuroscience
-
8/12/2019 Controlling Extraneous Variables Chapter5
11/20
Meaden, 1998), and this study indicated that bright light has an antidepressant
effect greater than the placebo effect.
There are ethical concerns when conducting a study using placebos. Participants
should be told before they give consent to participate in the study that they may
receive a placebo instead of an active treatment. In many instances, it is not ethicalto use a placebo. If there is an established treatment for the disorder, then the control
group should receive that conventional treatment. You then compare the effect of
your new treatment, given to the experimental group, to the effect of the conven-
tional treatment given to the control group. If you recruit more potential participants
than you are able to treat at one time, you can use the participants placed on the
waiting list as a control group, later providing these participants with a treatment.
SINGLE- AND DOUBLE-BLIND STUDIES
One way to control for placebo effects is to conduct a single-blind procedure.In a single-blind study, participants do not know if they are in the experimen-
tal or the control group. A single-blind study design helps to control for the par-
ticipants expectations about effects of the treatments you are testing. You
should carefully consider the ethical justification for such a design because par-
ticipants are not fully informed about exactly what treatment they will receive
before consenting to the experiment. It can be difficult to conduct a single-blind
study if the treatment produces an effect that is obvious to the participant. For
example, recall the cranial electrical stimulation discussed in Chapter 1. It
would be difficult to keep the participant blind in this instance because the elec-
trical stimulation produces a tingling sensation in the earlobes.
In a double-blind procedure, neither the participants nor the experimenters
administering the treatments know who is in the experimental group and who is
in the control group. Of course, in animal research, the animals do not have aware-
ness or expectancies about their assignment to experimental or control groups, so
we do not use the terms single-blindanddouble-blind. When an animal study is
conducted so that the experimenter is not aware of which subjects are in which
experimental group, we simply call it a blindstudy. Why conduct a blind study?
This design controls for experimenter effects,subtle effects the experimenter can
have on the outcome of the experiment. Generally, experimenter effects are biased
toward the hypothesis. This bias is in most cases unconscious and in many cases
quite subtle. For example, if you were conducting behavioral tests of mice that you
hypothesized were going to act more aggressive than the mice in the control group,
you might handle the mice differently as you placed them into the test arena, being
more hesitant in your handling of mice you expected to be more aggressive. These
mice might then act differently during the test, in part because of the difference in
handling. Another possibility would be that you would interpret an ambiguous
Chapter 5 Controlling Extraneous Variables 87
-
8/12/2019 Controlling Extraneous Variables Chapter5
12/20
behavior differently depending on whether you expected that mouse to be aggres-
sive or not. To protect yourself against the question of potential experimenter
effects, conduct your experiment blind to the treatments whenever possible.
Often researchers conduct ablind analysis, in which they code their samples so
that as they analyze the data they are unaware of prior treatments. For example, inanalyzing brain sections from people with autism to compare with brain sections from
controls, you could cover the original labels on the slides with labels that are coded.
During data collection, as you measure cell size or count cell numbers, you would be
blind to the identity of the brain sample, keeping track of your measurements using
the codes on the slides. Only after you finished collecting data would you look under
the labels and determine which slide came from which group of participants.
Striving for objective measures whenever possible will also protect you from
experimenter effects. Using an image analysis software package to determine cell
diameter is better than attempting to measure the diameter yourself, but both would
be better than judging if a cell is small, medium, or large. Using multiple
observers and checking for good inter-rater reliability is another good approach.Often experimenters will include a positive control and a negative control to check
that the procedure is working reliably. A positive control is a sample that should
definitely demonstrate a positive reaction, and a negative control is a sample that
should show a negative reaction. For example, in an experiment measuring cell
death in neural development, fragmented DNA would be one reasonable depen-
dent variable to measure, because this is associated with cell death. A positive
control would be a brain section to which the experimenter applied an enzyme that
will fragment the DNA (DNase). This specially treated brain section should give a
positive signal in the assay. A negative control might be a brain section that did not
receive one critical reagent but was otherwise treated identically to the other
sections. This section would not be expected to show staining in the assay.
Vehicle Controls and Sham Surgery
In animal research, one control group for a study involving drug administra-
tion is often a vehicle-control group. The vehicle is the solution in which the drug
is dissolved, such as saline or artificial cerebrospinal fluid. Animals in the vehicle-
control group are treated exactly as the animals getting the drug except that no
drug is dissolved in the vehicle. Thus, if the drug is administered with an intraperi-
toneal injection, the vehicle-control group animals receive the vehicle through an
intraperitoneal injection. Comparing results from the vehicle-control group with
a separate group of untreated animals allows you to detect any effects of the injec-
tion procedure on the dependent variable (see also Box 5.2).
If your experimental group is receiving surgical interventions, you
will want to include a control group receiving sham surgery. This group is
anesthetized for surgery, placed in any special apparatus necessary for
88 The Design of Experiments in Neuroscience
-
8/12/2019 Controlling Extraneous Variables Chapter5
13/20
surgery, and is treated exactly as the experimental group except for the step
that is thought to alter the critical variable.
Chapter 5 Controlling Extraneous Variables 89
BOX 5.2
TOOLS OF THE TRADEA Few Special Concerns for Pharmacological Studies
Several special concerns arise in the design of drug studies. It is critical to include a dose-response curve to char-
acterize effects, since these can provide information of the specificity and mechanism of action of the compound
being tested. A big mistake students often make is in the spacing of the doses administered. Generally, drug
doses are varied by log units (see Figure 5.4). An inadequate range of doses tested will not allow a full descrip-
tion of the dose-response relationship, which is the fundamental relationship to describe in initial studies.
The example in Figure 5.4 also highlights a common method for pooling results from cases that have
different baseline level responses. Here the researchers have normalizedthe responses to the maximum
current demonstrated by that particular oocyte in response to GABA. To do this, they took each measured
current and divided it by the maximum current for that particular ooycyte.
Figure 5.4 Dose Is Often Varied by Log Units
GABA
GABA + (+)-menthol
1.4
1.2
1.0
0.8
1 10
[GABA] M
Normalizedc
urrent
100 1000
0.6
0.4
0.2
0.0
A dose-response curve for oocytes treated with either GABA or with GABA and menthol, demonstrating the sensitivity of
GABA receptors to modulation by menthol. Note the spacing of drug concentrations used by log units, a common design
within pharmacological studies. Note also that the evoked current responses were normalized to the maximum response
for each oocyte tested to allow comparison across the sample.
SOURCE: Hall et al., 2004. Reprinted by permission of Elsevier.
-
8/12/2019 Controlling Extraneous Variables Chapter5
14/20
TIME-SERIES DESIGN
A time-series design is a within-subjects design that involves measures of the depen-
dent variable taken at different times, with the experimental or control treatment
applied between the measures. This is particularly suited to single-subject designs.
The simplest time-series design is the ABA design, which has three stages.
First, the dependent variable is measured under control conditions to establish a
CONTROLLING ORDER EFFECTS: COUNTERBALANCING
In the experiment described in Within-Subjects Designsection, where you were
measuring a dose-response curve for GABA action for each oocyte with and with-
out anesthetic, one potential worry is that you might observe an order effect if theorder of the experimental conditions influences the results. This is always a con-
cern for within-subject designs. For example, if you always tested the oocytes with
the anesthetic condition first, perhaps previous exposure to anesthetic would influ-
ence the GABA response in the control condition. On the other hand, if you
always tested the oocytes with the control condition first, then previous exposure
to GABA might influence the GABA response in the anesthetic condition.
What can you do to control for an order effect? Randomizing the order of the
conditions is one good approach. Another approach is to counterbalancethe order
of the conditions, to design the study so that equal numbers of subjects experience
the various possible orders of the conditions (see Figure 5.5). In our oocyte experi-
ment, this is relatively easy; we could test half the oocytes with the control conditionfirst and the anesthetic condition second; the other half of the oocytes would experi-
ence the anesthetic condition first and the control conditionsecond. Themain advan-
tage of using a counterbalanced order instead of a randomized order is that after using
a counterbalanced order, theexperimenter can measure the order effect and can deter-
mine if the order of conditions had a substantial effect on the response measured.
90 The Design of Experiments in Neuroscience
Figure 5.5 Counterbalancing for Order of Treatments in an Experiment
The four treatments, A, B, C, and D, are given to subjects so that every possible order is used and each of the four subjects
receives a different order of treatment.
Subject First treatment Second treatment Third treatment Fourth treatment
1 A B C D
2 D C A B
3 C D B A
4 B A C D
-
8/12/2019 Controlling Extraneous Variables Chapter5
15/20
baseline. Then, the experimental condition is applied and the dependent variable
is measured under the experimental conditions. Finally, there is a return to the
control condition and the dependent variable is measured again. This third con-
dition is important to include because a change in the dependent variable over time
might indicate an effect of the independent variable or might be simply from achange due to time. The third condition should show a return of the dependent
variable back to baseline levels. Thus, this design involves control (A), experi-
mental (B), and control (A) conditions (see Figure 5.6).
Chapter 5 Controlling Extraneous Variables 91
Figure 5.6 Time-Series Design
2enilesab1enilesab bright light experimentalRoom 2 (n=5)
bright light baseline 2experimentalRoom 1 (n=6) baseline 1
baseline 1 baseline 2bright light experimentalRoom 3 (n=3)
Exp. week: 21 222120191817161514131211109876543
)4(hcraM)3(yraurbeF)2(yraunaJ)1(rebmeceD:htnoM
Room 4a (n=4) bright light experimental
basebaseline 1
baseline 1 base
Room 4b (n=4) bright light experimental
(a)
(b)
This study examined the effect of bright light on the rest-activity rhythm of Alzheimer patients.(a) Protocol of the study: Patients were
ied during a baseline period and then exposed to bright light in the dayroom for 4 weeks, followed by a return to original condition
second baseline measure. The timing of the experimental treatment was staggered to allow measurement of a possible seasonal effeExample of data from one patient in the study: Activity was measured from a wrist-worn monitor. Five days of baseline activity before
ment (top panel) demonstrates the disorganization in daily rhythms associated with dementia. During bright light treatment, rest-a
rhythms become less variable with no change in amplitude (middle panel). After treatment is withdrawn, the patient reverts back to o
pattern (lower panel).
SOURCE: Van Someren, Kessler, Mirmiran, & Swaab, 1997.
-
8/12/2019 Controlling Extraneous Variables Chapter5
16/20
92 The Design of Experiments in Neuroscience
More complex time-series designs are possible. For example, you could have
an ABABA design, to determine if the effect of the independent variable replicates
over two applications. A time-series design with more than one independent vari-
able might be an ABCA design or an ABACA design.
All of these designs suffer from interpretational problems because thedependent variable may change over time independently of the experimental
conditions. Think of time as a major uncontrolled variable in these studies.
Necessarily, the measures taken under control and experimental conditions
vary by more than the independent variable; they also vary in the time
they were collected. Any other factor that has changed during that time
might alter your dependent measures and thus might account for any differ-
ences you observe. For example, in studies using fMRI activity as a depen-
dent variable, scanner drift, or noise intrinsic to the MRI machine
that slowly changes over time, can present problems in interpreting data
(see also Box 5.3).
BOX 5.3 TOOLS OF THE TRADE
The Design of fMRI Experiments
Special considerations apply to the design of experiments using fMRI to assess brain activity. The
textbookFunctional Magnetic Resonance Imaging(Huettel, Song, & McCarthy, 2008) describes them
well, and a student starting to work with fMRI experiments should consult it for more detailed
information.A common research design is the blocked design, where the experimental conditions occur in an
alternating order, with each level of the independent variable presented for an extended time, in
blocks (see Figure 5.7). This design is preferred when the experimenter is interested in changes in
brain state.
In some instances, an event-related designis preferable, where brain activity is measured following
short-duration presentations of stimuli in discrete events or trials (see Figure 5.8). This design would be
appropriate when an experimenter is interested in changes in the brain that are caused by properties of
presented stimuli.
Always in interpreting fMRI study results, keep in mind that you may be able to conclude that a change
in brain activity is associated with your task, perhaps that your task caused the change in brain activity,
but you can never conclude from studies using this technique that the brain activity caused the behav-ioral performance.
-
8/12/2019 Controlling Extraneous Variables Chapter5
17/20
Chapter 5 Controlling Extraneous Variables 93
Figure 5.7 Example of a Blocked Design in an fMRI Experiment
(a)
(b)
Carrot
Task A Task B
Task B
Task A Task B Task A Task B Task A Task B
(c)Task A
Mailbox
Pla nt Ha nd ba g P ebb le Ch ess Bo ok Ph on e An ge r Wat ch Wi ndo w Nig ht
Knife Tiger Sweater Teapot Auto Doorbell Spider Parsley
Rest Task ARest Rest Task B Rest
The participant reads 10 words with background music in one block (Task A).This contrasts with Task B, when the participant
reads ten words with no background music.The experimenter doesnt analyze the response to individual words, but assumes
that the cognitive processes of interest occur throughout the block.The experimenter can design the study as shown in (b) alter-
nating Task A and Task B blocks, or the experimenter can insert a rest condition between each block, as shown in (c).
SOURCE: Huettel et al., 2008. Reprinted by permission of Sinauer Associates, Inc.
Figure 5.8 Example of an Event-Related fMRI Experiment Design
Time
The participant simply watches a screen where occasionally a face or an object appears. The figure shows a time axis
and indicates the relative position in time when each image is presented. The experimenter analyzes the brain activity
immediately following the faces and compares this to fMRI activity following objects.
SOURCE: Huettel et al., 2008. Reprinted by permission of Sinauer Associates, Inc.
-
8/12/2019 Controlling Extraneous Variables Chapter5
18/20
CHECK YOUR UNDERSTANDING
1. Independent variable and dependent variable. For each experiment
below, define the independent variable and the dependent variable.
a. To determine if the pattern of brain activation differs when recall-
ing true versus false memories, researchers used PET scans to
observe brain activity when participants recalled words from a list
previously presented, as well as words very similar to the words on
the original list, ones the participants thought were on the list but
actually were not. The hippocampus was active in both instances,
but the left temporal parietal area was active only in the cases of
true memories.
b. A study measured responses from patients with mild brain injury
from a recent traumatic event with a control group matched on sev-
eral relevant variables. It measured selective and sustained attention,verbal and nonverbal fluency, and verbal memory (Mathias, Beall, &
Bigler, 2004).
c. The role of the molecule Hedgehog in brain development was
investigated using zebrafish. In the first experiment, Hedgehog sig-
naling pathways were pharmacologically blocked, and researchers
observed that development of the pituitary was disrupted. In the sec-
ond experiment, researchers reported that zebrafish with genetic
mutations in the components of the Hedgehog signaling pathway
showed alterations in the structure of the pituitary (Sbrogna, Barresi,
& Karlstrom, 2003).
2. This experiment has a major flaw. What is it?Does luteinizing hormone
releasing hormone (LHRH) increase the number of calcium channels in
neurons? To test this, a researcher applied LHRH to cell cultures and
measured the density of calcium channels. While there was a progressive
loss in cells, an expected result for this type of cell culture, the surviving
cells showed increased numbers of calcium channels. A control group of
cell cultures without LHRH did not show the cell loss or increased
numbers of calcium channels. The researcher concluded that results sup-
ported the hypothesis that LHRH increased the number of calcium
channels.
a. No control groupb. A major confounded variable
c. Selective attrition
d. Experimenter bias
94 The Design of Experiments in Neuroscience
-
8/12/2019 Controlling Extraneous Variables Chapter5
19/20
THINK ABOUT IT
Each of the experiments described below has a major flaw. Can you determine
what it is?
1. A researcher wanted to demonstrate that the use of Ritalin in childhood
predisposed women to substance abuse in adolescence. She asked all herfriends how many years they had taken Ritalin as children and how
often they took drugs in college. She had to discard several data points
because they were probably not accurate; she was pretty sure some
friends had not told her the truth. Her results supported her hypothesis.
2. In an experiment using antibodies to label serotonin-containing neurons,
the brain sections processed with no antibodies applied to them are:
a. Positive control
b. Primary control
c. Negative control
d. Inverse control
3. From a well-controlled fMRI study showing that when participants
judged a moral conflict there was increased activity in the Anterior
Cingulate Cortex (ACC) you can conclude that:
a. Brain activity in the ACC caused the moral choice.
b. Considering moral conflicts caused a change in activity in the ACC.
c. People judging moral conflicts are less likely to have increased activ-
ity in the ACC.
d. The ACC is a brain area controlling moral conflict resolution.
4. Show how to counterbalance the order of treatments A, B, C, and D.
Chapter 5 Controlling Extraneous Variables 95
SubjectFirst
treatmentSecond
treatmentThird
treatmentFourth
treatment
1
2
3
4
-
8/12/2019 Controlling Extraneous Variables Chapter5
20/20
2. A student was under time pressure to get his thesis completed.
Unfortunately, the drug he needed for his experiment was on back order
and would not be available for another month. He used the month to
collect all his control data, so that when the drug arrived, all he had to
do was to collect the data from the drug-treated group to complete hisexperiment.
3. To demonstrate the role of integrins (cell adhesion receptors) in changes
in the nervous system associated with learning and memory, a research
team measured several changes in the hippocampus that were well
established as measures related to learning and memory. They compared
results from control experiments using mice from a common inbred
strain to results from knockout mice missing a specific integrin.
4. When Josie counted labeled cells in Area X, there were, as she had
expected, very few labeled cells, or at least the cells were not so densely
labeled that she thought they were appreciably above background levels.When she moved to count cells in Area Z, much to her delight, her
hypothesis was confirmed: these cells looked much more densely labeled.
SUGGESTIONS FOR FURTHER READING
Bandettini, P. A. (2009). Whats new in neuroimaging methods? Annals of the New YorkAcademy of Science, 1156, 260293.
Gavriaux-Ruff, C., & Kieffer, B. L. (2007). Conditional gene targeting in the mouse ner-vous system: Insights into brain function and diseases.Pharmacology Therapy, 113,619634.
McCutcheon, J. E., & Marinelli, M. (2009). Age matters. European Journal ofNeuroscience, 29, 9971014.
Pangalos, M. N., Schechter, L. E., & Hurko, O. (2007). Drug development for CNS dis-orders: Strategies for balancing risk and reducing attrition.Nature Reviews DrugDiscovery, 6, 521532.
Pollo, A, & Benedetti, F. (2009). The placebo response: Neurobiological and clinicalissues of neurobiological relevance.Progress in Brain Research, 175, 283294.
96 The Design of Experiments in Neuroscience